We show our math.

For each brand we generate buyer-intent questions across families: best-of / recommendation, comparison, alternatives, use-case, budget, and trust. These are the questions real buyers ask, not branded vanity queries. You can edit, add, or remove any of them.

We query the AI tools buyers actually use (ChatGPT, Claude, Gemini, and Perplexity, plus Grok on Grow and Agency plans) and track each in two distinct modes:

• Model memory: what the model says from training alone, no live search.
• Live search (grounded): the model with its web-search tool on, which is what most users now experience.

These often disagree, and conflating them hides where you are actually winning or losing. We label every answer with its mode. Every paid plan includes every AI tool. No per-model upcharges.

Because answers vary run to run, a single query is noise. On paid plans we ask each question multiple times per AI tool and mode, then report the aggregate, plus a stability figure showing how consistent the answers were. Low stability is a signal to trust the number less, and we show it rather than hide it.

We only record a numeric rank when the answer contains an explicit ordered recommendation list that includes your brand. A brand mentioned in prose is a mention, not a rank. Our headline positioning metric is share of voice and top-3 rate across many samples, aggregates that are meaningfully stable, rather than a single ordinal position, which is not.

A language model reads every answer with a strict extraction schema: does it mention your brand (including misspellings and product-line names, and disambiguating your brand from unrelated words that happen to match), at what rank if any, with what sentiment, citing which sources, and with any factual errors about you. Truncated answers, provider errors, and empty answers are stored but excluded from metrics, so a cut-off list never corrupts a rank and a blank response never counts as “not mentioned”. Every report says how many valid answers it is based on and how many were excluded.

The 0–100 visibility score is a weighted blend of mention rate, rank quality, citation rate, competitive pressure, and sentiment. We always show the raw components next to the score so it is never a black box. The exact weighting is documented in-product and may be refined; when it changes, historical runs are recomputed so trends stay comparable.

Every recommended fix carries the exact prompts it targets. When you mark it shipped, we freeze the mention rate on those prompts and re-measure them on every later full audit. The rest of your tracked prompts get the same numbers and serve as a control group, because AI answers drift on their own. A fix only counts as a win when its targeted prompts move more than the control; when the two move together, the receipt says so instead of claiming credit.

• AI answers change constantly. Treat week-to-week movement inside the stability band as directional, not precise. We label it that way.
• We measure a defined question set, not the infinite space of phrasings. Broader coverage means more questions, which higher plans allow.
• Personalization, region, and model version affect what any individual user sees. We query neutrally and report the aggregate.
• Visibility is a leading indicator, not revenue. We tie it to referral signals and to the actions you ship, but we will not pretend a citation is a customer.